A popular model of single microlensing light curves neglects the possi
bility that only a fraction of the light is due to the lensed star, th
e remaining part being due to a close, unresolved blend, which may be
related or unrelated to the lens. Unfortunately, the effects of blendi
ng are significant, as all microlensing experiments choose very crowde
d fields as their targets. The propensity of blends among double lense
s is the most direct evidence for the seriousness of the problem. In t
his paper, we point out a strong degeneracy of the fitting procedure f
or single lensing events, which makes it very difficult to detect the
presence of a blend, and practically impossible to correct for it by p
urely photometric means, in a large part of the parameter space. Some
blends may be detected by astrometric means, but the majority have to
be corrected for statistically. It will be helpful to measure the lumi
nosity function well below the groundbased detection limit, e.g., usin
g the Hubble Space Telescope (HST). Binary stars in the target populat
ion affect blending and may be undetectable with HST; their presence w
ill be revealed by repeating microlensing events, so it will be possib
le to correct for them statistically. If no correction is made, then t
he event timescales and the lens masses are systematically underestima
ted.